Forthcoming, Journal of Public Economics
Social Preferences, Self-Interest, and the Demand for Redistribution
Christina Fong*
July 27, 2000
Abstract
Preferences for redistribution may be influenced by values and beliefs about distributive
justice as well as by self-interest. People may prefer more redistribution to the poor if
they believe that poverty is caused by circumstances beyond individual control.
Therefore, beliefs about the causes of income may affect demand for redistribution.
Alternatively, the effect of these beliefs on redistributive preferences may be spurious if
they are correlated with income, and self-interest is not properly controlled for. They
may also measure incentive cost concerns. Using social survey data, I find that selfinterest cannot explain the effect of these beliefs on redistributive preferences.
Keywords: Equity, Fairness, Interdependent Preferences, Redistribution, Distributive
Justice
JEL Classifications: D63, H50
*
Send correspondence to Christina Fong, Department of Economics, Thompson Hall, University of
Massachusetts, Amherst, MA 01003. Tel: (314) 633-1163. Email: [email protected].
1
1.
Introduction
The reasons why citizens of democratic countries support or oppose
redistribution to strangers remain poorly understood, despite much research on the public
sector and welfare states. Paul Romer (1994) has observed that social security in the U.S.
was designed and implemented as a program of entitlements to those who have
contributed their earnings rather than as a program that would provide broader coverage
and hence redistribution to those who had not contributed. According to Romer the
designers and subsequent defenders of social security understood that a program of
entitlements would inspire more vigorous public support than the alternative.
Why might voters defend entitlement programs more vigorously than other kinds
of programs? One possible explanation is that social security is a social insurance
program that appeals primarily to self-interested desires on the part of a broad range of
citizens. A self-interested demand for social insurance can arise where private provision is
prevented by market failures that result from asymmetric information (Sinn, 1995; Barr,
1992). Tying entitlements to contributions may simply be a method of controlling
potential costs due to moral hazard. Voters might desire social insurance, but only if
certain measures are in place to prevent people from working less after they become
insured.
On the other hand, the appeal of a program of entitlements may have to do with
preferences that are other-regarding and not necessarily outcome oriented. These are
often referred to as social preferences. Romer (1994) argues that if people feel entitled to
a benefit, they may feel angry if they do not receive it. Such anger might be expected if
people care deeply that their opinions about what they deserve are realized, but a
financially self-interested rational actor should not recognize the difference between an
entitlement and a handout. Following many others, I will argue that individuals may also
care deeply that other people get what they deserve. Many scholars have argued that
people may hold the principle of justice known as equity, according to which the rewards
or incomes that people receive should be in proportion to the contributions they make
(Walster, Walster and Berscheid, 1978; Deutsch, 1985).
Reciprocity may produce related concerns. (Bowles and Gintis, 2000) Wealthy
people may desire a society in which members who are well-off support those in need,
provided that others would do the same for them if necessary. Such reciprocity might
require that those in need are predisposed to work hard: a willing tax payer might well
withdraw his or her support for redistribution if recipients are thought to be voluntarily
capitalizing on tax payer generosity.1 Other motivations such as altruism, either to kin or
1
There is a large body of literature citing evidence that reciprocity is a strong human motive that governs
the allocation of resources according to how recipients have behaved or intended to behave in the past.
Most of this literature concerns reciprocity in bargaining or strategic interactions among small numbers of
people, but Bowles and Gintis (2000) argue that a generalized form of reciprocity applies to large scale
redistribution.
2
non-kin, may explain support for redistribution.2 A third possibility is that respondents
may support redistribution to prevent crime or other forms of social unrest.
There is strong evidence that financial self-interest is an insufficient explanation
for redistributive attitudes. First, means-tested programs which benefit a relatively small
range of people have proven politically viable, albeit at lower levels of political and fiscal
support. Second, social survey data suggest high levels of support for certain
redistributive programs, even among the rich. Surveying a variety of American public
opinion data sources, Martin Gilens (1999) reports that roughly fifty and sixty per cent
of respondents express support for increases in government spending on Medicaid and
child care for poor children respectively. In contrast, less than 10% of respondents
express support for decreases in government spending on these programs. Further
evidence - from the 1998 Gallup Poll Social Audit - is that there is substantial support for
redistribution among those who are unlikely to benefit from it. Among respondents who
have annual household incomes of at least $150,000 and expect to be upwardly mobile in
the next five years, a sizeable fraction respond in favor of reducing inequality or helping
the poor. In this group, 24% respond that the government should “redistribute wealth by
heavy taxes on the rich,” and 67% respond that the “government in Washington DC
should make every possible effort to improve the social and economic position of the
poor.” (See Appendix) Equally striking is the fact that among those with annual family
incomes of less than $10,000, 35% report that the government should not redistributive
wealth by heavy taxes on the rich, and 21% say that the poor should help themselves
rather than having the government “make every possible effort to improve the … position
of the poor.”
A number of authors have argued, both theoretically and empirically, that
redistributive preferences may be determined by beliefs about the extent to which
individuals do or do not have voluntary control over their earnings, and beliefs about the
prevalence of opportunity to get ahead.3 (Kluegel and Smith, 1986; Piketty, 1995; Gilens,
1999) These beliefs about the causes of poverty and wealth are usually arranged according
to beliefs about self- versus exogenous-determination of outcomes. Those who believe in
self-determination believe that outcomes are determined by factors that are within
individual control, such as a willingness to work hard. Those who believe in exogenousdetermination place more importance on factors beyond individual control, such as luck or
lack of opportunity.
Many authors across the social sciences have interpreted the effect of self- and
exogenous determination beliefs on redistributive attitudes as reflecting reciprocity
2
There is a long tradition of modeling public redistribution with altruistic preferences. For a recent
application, see Moffitt (1998, 1999).
3
Standard median voter models of both altruistic and purely self-interested demand for public
redistribution are easily extended to accommodate this finding. See, for instance, Luttmer (1998). A
second argument for extending models of simple altruism, from the literature on charitable giving, is that
they fail empirical tests of Ricardian equivalence. That is, if altruism is simply an argument in the donor’s
utility function for the incomes of the recipients or the total contribution to a public good, then
government transfers should completely crowd out private giving to the poor. Sugden (1988) provides
evidence that this is not the case. Andreoni (1989) provides a model of warm-glow altruism in which
individuals also care about the size of their contribution to recipients or a public good.
3
(Bowles and Gintis, 2000) or the principle of equity (Kluegel and Smith, 1986).4 In both
cases, the interpretation is that people are willing to support the poor conditional on
them having industrious traits or intentions.5 However, it is possible that the effect of
self- and exogenous-determination beliefs is consistent with financial self-interest. People
who believe in exogenous-determination may be those who have low-mean, high-variance
incomes. Such individuals may have higher expectations of needing government assistance
in the future, and therefore demand more redistribution. For similar reasons, those who
believe in self-determination may simply be people who have higher-mean, lower-variance
incomes and therefore less self-interest in redistribution.6
This paper tests various long-standing hypotheses about why people support or
oppose redistribution, but the primary focus is on the role of beliefs about self- and
exogenous-determination on reported redistributive policy preferences. If the beliefs
about self- and exogenous-determination operate through self-interest, then they should
have no effect among people who do not expect to gain financially from redistribution. I
will first test whether this is the case using a sub-sample of individuals with high incomes
who expect to be upwardly mobile in the next five years. In this sub-sample, a large set
of self-interest measures is jointly insignificant, according to a likelihood ratio test. I test
whether or not beliefs about self- and exogenous-determination are significant in this subsample. I then conduct an analogous test on the sub-sample of people with annual family
incomes of less than $10,000. The self-interest variables are jointly insignificant in this
sub-sample. I test whether or not beliefs about self- and exogenous determination are
jointly significant at the one percent level.
I also test whether beliefs about the prevalence of poverty affect attitudes about
redistribution. If self-interest fully explains demand for redistribution, then those who
have no self-interest in redistribution should be opposed to it regardless of their beliefs
about the prevalence of poverty. If these beliefs do have a significant effect in a highincome, upwardly mobile sub-sample, then an alternative explanation is necessary.
In my fourth test of self-interest, I use questions on the importance of various
factors – including a person’s race and gender - to getting ahead in life. If people think
that a person’s race and gender are important to getting ahead in life, then effects of these
beliefs on self-interested demand for redistribution should operate in opposite directions
for those who expect to benefit and those who expect to lose from racial or gender
discrimination.7
2.
Data
4
See Miller (1985) for a survey of empirical research on distributive justice.
There is evidence from experimental bargaining games that people may care directly about intentions (see,
for instance, Blount 1995), but the role of beliefs about intentions in redistributive politics is unknown.
6
Another way in which beliefs about self- and exogenous-determination might operate through self-interest
is if they represent concerns about incentive costs of taxation and redistribution. I find evidence against
this argument using the U.S. General Social Survey. I discuss the argument and the evidence against it in
the conclusion.
7
I assume that people agree on which group benefits and which loses when they believe that a person’s
race or gender is important to getting ahead.
5
4
I use the 1998 Gallup Poll Social Audit Survey, “Haves and Have-Nots:
Perceptions of Fairness and Opportunity,” a randomly selected national sample of 5001
respondents. There are 3626 individuals who report being in the labor force. In each test,
I use the set of all individuals in the labor force who responded to all of the questions
used in the regression, unless noted otherwise.8 Several tests employ a sample of 2738 or
high-income subsets of this sample.
The Gallup Poll Social Audit has a large sample size for a large number of
questions on inequality and distribution relative to other commonly used surveys. The
sample size permits running regressions with full controls on narrow segments of the
sample, namely, high income and low income sub-samples. There is a large number of
self-interest measures that include not only the usual objective socioeconomic variables,
but also subjective measures of economic well-being which may widen the net intended to
capture self-interest.
The wording of selected questions used in this study is provided in the appendix.
The dependent variable is a summative scale of five questions on whether or not
inequality should be reduced, either by the government, by heavy taxation of the rich, or
in general. Questions about the fairness of the distribution of income and wealth that do
not specifically ask about reducing inequality or helping the poor are available but not
used in order to reduce the chances of measuring unrelated concepts.
Two sets of measures of beliefs about self- and exogenous-determination of
incomes are used in this study. The first contains two questions concerning the
importance of effort and luck in causing wealth and poverty, and one question on whether
or not there is plenty of opportunity to work hard and get ahead in America today. The
second set is a series of questions about the importance of various factors – including race
and gender - for getting ahead in life (see Appendix).
Self-interest is measured with the standard objective socioeconomic controls: race,
gender, income, education, marital status, age, employment status, union status,
occupation, and household size.9 In addition to these, I use subjective measures of
economic well-being, namely the frequency with which respondents worry about meeting
family expenses, and a measure of future mobility.10 All of these variables may affect
redistributive preferences through expected future government receipts.
To measure the respondents’ beliefs about the prevalence of poverty, I use a
question about whether or not the respondent thinks that America is a society of “haves”
8
I drop non-responses and “don’t know” responses. Another option would be to include “don’t know” as
a valid response. However, how and why people develop well-defined preferences and beliefs is beyond the
scope of this paper. I focus on why people oppose or support income redistribution given that their beliefs
and preferences are well defined.
9
Some of these variables, especially union membership, are not direct measures of financial self-interest,
but correlate with income, and are intended as proxies for self-interest.
10
There are several additional questions that might capture self-interest that I exclude from the model. I
test formally for their joint significance. Using a likelihood ratio test, I fail to reject the hypothesis that
the coefficients of these variables are all zero (at the 10% significance level) in an ordered probit regression
like that presented in Table 2 below. This test may be interpreted as a test for missing variables. After
dropping these variables from the analysis, the sample size increases by over 400 observations, allowing
me to increase the income cutoff for the high-income sample.
5
and “have-nots.” Strictly speaking, this is a measure of beliefs about the prevalence of
inequality as well as poverty, but it is the best measure in data set.
3.
Empirical Approach
I begin with an ordered probit estimate of a basic equation predicting support for
redistribution with beliefs about self- and exogenous-determination and the self-interest
measures. I use the full sample of 2738 individuals who responded to all of the questions
included in the model.
Let R* = XBβB + XSβS + _ be the latent regression where R* is an unmeasured
level of support for redistribution, XB is a matrix of variables on beliefs about self- and
exogenous determination, XS is a matrix of self-interest measures, and βB and βS are
vectors of coefficients. The actual measure of support for redistribution has six ordered
categories, i = 1 to 6. The probabilities of each response category are:
Pr(R = 1) = Pr(XBβB + XSβS ≤ k1)
Pr(R = i) = Pr(ki- 1< XBβB + XSβS ≤ ki) for R = 2,…,5
Pr(R = 6) = Pr(ki-1 < XBβB + XSβS)
(1)
where k1, k2, … k5 are the cutpoints and _ ∼N(0,1) is the error term.
I use the likelihood ratio test to test for the joint significance of each subset of
coefficients. That is, I test:
H0: R* = XBβB
+_
HA: R* = XBβB+XSβS +_
(H1)
H0: R* =
XSβS +_
HA: R* = XBβB+XSβS +_
(H2)
I present the marginal effects of selected variables from the model.11 The marginal
effect on category i of R for a dummy variable D is Pr[R = i] evaluated at D = 1 and the
mean of the remaining regressors minus Pr[R = i] evaluated at D = 0 and the mean of the
remaining regressors.12
Next, using the same specification, I use sub-samples of individuals who are so
well-off and secure about their futures that self-interest measures are jointly insignificant,
11
Note that in ordered probit, the effect of coefficients is to shift the probability distribution of R* and R.
The sign of the coefficient will be the same as the effect on the highest categories of the dependent variable
and opposite of the effect on the lowest category. The direction of the marginal effects on the remaining
values of the dependent variable cannot be inferred from the coefficient alone.
12
The predicted probabilities of the categories of R are calculated as follows:
Prob(R = 1) = Φ (k1 - XB)
Prob(R = i) = Φ (ki - XB) -Φ (ki - 1 - XB), for i = 2,…,5
Prob(R = 6) = 1 - Φ (k5 - XB),
where Φ is the cumulative standard normal.
6
according to the likelihood ratio test of (H1). The results I present are from a sub-sample
of 333 individuals who have annual family incomes of $75,000 or more, expect to their
lives to improve in the next five years, and worry about their bills less often than “all of
the time.”13 I then conduct (H2) using the likelihood ratio test. If we reject the null
hypothesis in (H2), but do not reject the null hypothesis in (H1), then some explanation
other than financial self-interest is needed.
I repeat these tests on the low-income sample. The only difference is that I
include individuals who are not in the labor force in order to have a sufficiently large
sample. In order to do this, I drop occupation from the equation, because the nonresponse rate among people who are not in the labor force is high for this question. The
sample contains 176 individuals with incomes under $10,000 per year who worry about
bills more often than “almost never.”
I test whether or not beliefs about the extent of poverty have a significant effect
on support for redistribution in the high-income sample described above. If self-interest
explains demand for redistribution, then individuals who cannot plausibly expect to
benefit from redistribution should be opposed to it regardless of their beliefs about the
extent of poverty. Another reason for using the high income sub-sample is that beliefs
about whether or not the U.S. is a society of haves and have-nots are correlated with the
self-interest measures. A measure of perceptions of poverty may simply serve as a
proxy for self-interest if it is highly correlated with the individual’s own expected income
and the expected financial benefits of social insurance.
In a different specification, I conduct a test of the self-interest hypotheses using
measures of the importance of a person’s race and gender to getting ahead in life. These
beliefs may have two effects among purely financially self-interested people. First, if
people think men benefit when gender is important to getting ahead, then men who hold
this belief should have lower expectations of benefiting from redistribution, and women
who hold this belief should have higher expectations of benefiting. This effect should
increase support for redistribution in women who think that gender is important to
getting ahead, and decrease support in self-interested men who think gender is important
to getting ahead. This reasoning also applies to race and beliefs that a person’s race is
important to getting ahead. That is, the interaction term between the belief that gender is
important to getting ahead in life and a dummy variable for male should be significant and
negative if people are purely self-interested, and analogously for race and beliefs about the
effect of race on getting ahead.
3.
Results
Selected summary statistics for the full sample of 2738 individuals are presented
in Table 1. A large majority (83%) of the respondents believe that there is plenty of
opportunity to work hard and get ahead in America today. Forty seven percent of the
13
The results are not sensitive to the sample choice. The beliefs variables are jointly significant in every
high-income sample that I tried, including one of individuals who reported annual family incomes of
$100,000 or more and expected their lives to improve in the next five years.
7
respondents report that lack of effort is to blame for poverty, while 57% believe that
effort is responsible if a person is rich. (See Table A2 for question wording.) A majority
of the respondents report support for redistribution on four of the five questions used to
construct the dependent variable. (See Appendix)
Thirty-eight percent of the
respondents respond that America is a society of haves and have-nots. Compared to the
full sample, the high-income, upwardly mobile sub-sample has stronger average beliefs in
self-determination, and has a higher percentage of people who are white, male, college
educated, non-union, and married. Fewer people in this group worry about bills all of the
time, and more of them have a full-time job.
In column 1 of Table 2 I present results from the basic ordered probit regression
of support for redistribution on beliefs about self- and exogenous-determination and the
measures of self-interest. The coefficients on beliefs about self- and exogenousdetermination all have the expected signs. Those who believe that wealth and poverty are
caused by external circumstances and those who believe that both external circumstances
and effort level are important show more support for redistribution than those who
believe that effort level alone determines whether a person is rich or poor. The five beliefs
dummy variables are both jointly and individually significant at the one percent level.
All of the significant self-interest variables have effects in the expected direction.
Those with annual family incomes of $150,000 or more are significantly less supportive
of redistribution than those with incomes under $10,000. Whites, men, those with some
college education or more, married people, non-union members, are all significantly less
supportive of redistribution than their counterparts. Those who worry about bills some
of the time and those who almost never worry about bills are significantly less supportive
of redistribution than those who worry all of the time. The self-interest variables are
jointly significant at the one percent level.
In order to offer some interpretation of the basic regression (Column 1) in Table 2,
I present selected marginal effects in Table 3. If we ignore the bold face type for a
moment, then the columns of Table 3 contain the estimated probabilities that different
types of people will respond with a particular value of support for redistribution. There
are six categories of the dependent variable, labeled 1 for the lowest level of support for
redistribution to 6 for the highest. Someone who believes that bad luck causes poverty
has an almost zero probability of reporting the lowest level of support for redistribution,
and a probability of .25 of reporting the highest level of support. The marginal effects are
presented in the rows labeled “change” in bold face type. These are the differences
between two types of people in the estimated response probabilities.
From this table, the magnitudes of the beliefs variables are apparent. In every
column, the marginal effects of reporting that bad luck as opposed to lack of effort causes
poverty are roughly 2 times or more in magnitude than the marginal effects of being white
instead of black, male instead of female, and having some college education or more as
opposed to having no college education. The marginal effects of having an income of
$150,000 or more rather than an income of less than $10,000 are also smaller than the
marginal effects of the self- and exogenous determination beliefs.
8
In Figure 1 I present predicted probabilities of responding in each category of
support for redistribution for four extreme types of people. In this illustration, I choose
the different types of people to maximize the differences in the estimated probabilities
between them,14 subject to either the beliefs variables or the self-interest variables being
held at the sample mean. The first bar in each of the six categories is the predicted
probability for someone with the strongest beliefs in exogenous-determination who has
average values for all of the self-interest measures. This is someone who believes that bad
luck causes poverty, good luck causes wealth, and that the US is not a land of
opportunity, and on all of the other variables has the sample mean. The second bar in
each category of support for redistribution is someone who believes that lack of effort
causes poverty, strong effort causes wealth, that the U.S. is a land of opportunity, and
who has the sample average on all of the other variables. The difference in the height of
these two bars is the marginal effect of being one of these types instead of the other.
The third bar in each category of support for redistribution is someone with
average beliefs about self- and exogenous-determination, but who is in the lowest socioeconomic category. This person is non-white, female, single, a union member, and a parttime worker.15 She has no college education, has an annual family income of less than
$10,000, has a household size that is greater than four, and reports that she almost always
worries about bills. Her occupation is calculated according to the distribution of nonmanagement or non-finance occupations in the sample. The fourth bar in each category of
the dependent variable is again someone with average beliefs about external constraints
and self-determination, but who is in the highest socio-economic category. This person is
white, male, married, non-union, a full-time worker in management or finance, has a
household size of less than four, and almost never worries about bills. He has some
college education or more, and has an annual family income of $150,000 or more. What
Figure 1 suggests is that the effect of self- and exogenous-determination beliefs is almost
as large as the effect of being extremely privileged compared to being extremely
underprivileged.
In column 2 of Table 2, I present ordered probit coefficients from the basic
equation for a high-income, upwardly mobile sub-sample. The sub-sample in column 2
has an annual family income of $75,000 or more, expects life to improve, and worries
about bills less frequently than “all of the time.” I present results from the likelihood
ratio tests of (H1) and (H2). The self-interest variables are jointly insignificant while the
self- and exogenous-determination beliefs are jointly significant at the 1 percent level. The
key point of this test is that in samples that are so well-off that there is no plausible or
statistical effect of self-interest on support for redistribution, beliefs about the causes of
income still have a large and highly significant effect.
I find similar results (not reported) from the likelihood ratio tests of (H1) and
(H2) on the sub-sample of respondents who have annual family incomes of less than
$10,000 and who worry about bills more often than “almost never.” The self-interest
14
The single exception is that I use the sample mean of age and age squared for all four types of people.
The category “unemployed” did not have a significantly different effect from full-time employees, so the
comparison that shows the largest self-interest effect is between full-time and part-time workers.
15
9
measures are jointly insignificant at the 10% level.16 The beliefs in self- and exogenousdetermination are jointly significant at the 1% level and have the expected signs.
In column 3 of Table 2 I add a dummy variable for whether or not America is a
society of haves and have-nots to the specification. This variable has a significant
positive effect on support for redistribution. To the extent that this variable measures
beliefs about the prevalence of poverty, or indicates the respondent’s perception of the
size of the poverty problem in the United States, this result is inconsistent with selfinterest. High-income, upwardly mobile individuals motivated purely by financial selfinterest should not demand redistribution, and therefore the prevalence of poverty should
be irrelevant to them. One possible interpretation is that the effect of beliefs about the
extent of inequality in the U.S. has something to do with the perceived need for
redistribution: to maintain constant benefits, redistributive spending has to be higher if the
perceived number of potential recipients is higher.
Table 5 presents a regression of support for redistribution on the importance of
various determinants of success, controlling for the full set of self-interest measures. The
beliefs about the importance of the various factors to getting ahead in life are all
significant, except for beliefs about the importance of talent. The significant effects are in
the expected directions. The negative finding on talent is consistent with previous
research.
The results in Table 5 are difficult to explain by self-interest. If the effect of these
variables on support for redistribution operates solely through self-interest, then beliefs
that race and gender determine success should result in a negative interaction term for men
and whites because of their lower expected government receipts. Yet, the interaction
terms show that there is no significant difference between the slope coefficients of men
and women on beliefs that a person’s gender is important to getting ahead. The findings
on race are even stronger. The effect of believing that race is important to getting ahead in
life is significantly more positive for men than for women. This interaction effect is in the
opposite direction from that predicted by self-interest.17
4.
Conclusion
Many scholars argue that self-interest is a sufficient assumption to make in the
study of redistributive politics, and some still strongly defend this position. To others, it
is equally obvious that social preferences play an important role. This paper attempts to
offer a few rigorous tests on the character of redistributive preferences.
16
The self-interest measures included in this equation are race, gender, age, age squared, the frequency with
which the respondent worries about bills, a dummy for household size greater than four, a dummy for parttime workers, and a dummy for those who are not working. The not-working dummy includes the
unemployed as well as people who are not in the labor force.
17
The findings concerning the interactions of race and gender and beliefs about the importance of a
person’s race and gender to getting ahead are quite stable with respect to specification and sample size
changes.
10
From the perspective of the self-interest model, income is a surprisingly poor
predictor of redistributive beliefs. The marginal effects of having an income of $150,000
or more, compared to an income of under $10,000 are smaller than the marginal effects of
each of the three measures of beliefs about the role of effort, luck, and opportunity in life
outcomes. In fact, even the combined effect of a wide set of possible self-interest
measures is roughly equivalent to the combined effect of the three beliefs measures.
Furthermore, there is variance in the redistributive attitudes of the richest and poorest
people in the sample, and this needs to be explained. In the high and low income samples,
self-interest measures are jointly insignificant. In contrast, beliefs about self- and
exogenous determination are strong predictors of support for redistribution in the full,
high-income, and low-income samples.
However, before we conclude that these results are evidence of reciprocity or
equity, let us consider another mechanism through which self- and exogenousdetermination beliefs may affect demand for redistribution: through the perceptions of
and concerns about the incentive effects of taxation. In theory, when earned income is
more sensitive to work effort, taxation could cause greater effort disincentives. That is, tax
costs may be increasing in the income elasticity with respect to effort. If so, then beliefs
about the roles of effort, luck, and opportunity in generating income may affect the level
of support for redistribution. (Piketty, 1995) This type of incentive concern need not
apply only to self-interested demand for redistribution. Tax cost concerns can reduce
demand for any tax-funded expenditure, including redistribution that arises out of selfinterest, altruism, the equity principle, reciprocity, or unrelated expenditures such as
national defense. (Atkinson, 1999)
I use the 1990 General Social Survey to produce some evidence on this matter. I
estimate ordered probit regressions predicting support for spending on welfare, national
defense, halting the rising crime rate, and dealing with drug addiction, respectively.18 (See
Appendix) The independent variables are beliefs that the poor are poor because of lack of
effort, and five demographic variables (income, education, race, gender, and age). Two
self-determination belief dummies - lack of effort is very important and lack of effort is
somewhat important in causing poverty – have negative effects on support for
redistribution (significant at the 1% level) relative to believing that lack of effort is not
important. However, beliefs in self-determination have no effect on support for spending
on crime or drug addiction, and they have a significant positive effect on support for
spending on defense.19 If the self-determination beliefs simply measure tax cost concerns,
then their effect on support for all of these expenditure items should have been negative.
18
The sample size in these regressions ranges from 584 to 594.
There is a second type of potential incentive cost that I will call transfer incentive costs. This is the
possibility that transfer payments might affect the labor market decisions of recipients. (Atkinson, 1999)
It is unclear why transfer incentive costs would depend on the income elasticity with respect to effort.
However, it is possible to construct theoretical examples where it does, so I offer some suggestive evidence
from the 1990 General Social Survey. Suppose we accept the evidence above against tax cost concerns and
test whether or not self- and exogenous-determination beliefs capture the effect of transfer costs only. I use
ordered probit to estimate a model that predicts support for spending on welfare, foreign aid, on
“improving the conditions of blacks,” controlling for beliefs about self- and exogenous-determination and
five demographic variables (education, income, race, gender, and age). Those who think poverty is caused
19
11
The finding that self-determination beliefs have no effect on support for spending
on crime or drugs relates to the question of whether or not prudential concerns about
controlling social unrest explain redistributive attitudes. The GSS evidence suggests that
the effect of self-determination beliefs on support for redistribution cannot be explained
by tax cost concerns combined with prudential motives. However, we can test for the
direct effect of prudential concerns. I estimated an ordered probit model predicting
support for welfare spending. The independent variables are beliefs in self-determination,
support for spending on crime, and five demographic variables (education, income, race,
gender, and age).20 I also estimated an equation that is identical except that it included
support for spending on drug addiction instead of support for spending on crime.21
Support for spending on both crime and drugs has a significant positive effect on support
for spending on welfare. These effects are slightly smaller and less significant than the
effect of self-determination beliefs. This can be interpreted as evidence that prudential
concerns for controlling crime affect support for redistribution, along with self- and
exogenous-determination beliefs, and financial self-interest. Unfortunately, the sample
sizes are too small to check for robustness of this result with respect to the kinds of
specification and sample changes that are presented for the Gallup data.
The effect of beliefs about the prevalence of poverty on support for redistribution
may be interpreted in several ways. First, the belief about the prevalence of poverty may
be a perception of the number of potential recipients. To hold benefits constant,
spending must increase if there are more recipients. Given this, any explanation that
could explain support for redistribution in the high-income sample would be consistent
with this finding. In unreported regressions, I found that the belief about the prevalence
of poverty is usually significant whether or not we control for self- and exogenousdetermination beliefs, but it is not as robust to sample size and specification changes as
the self- and exogenous-determination beliefs. Therefore, whether the findings are
interpreted as consistent with altruism or equity, the strongest result in this paper is the
effect of self- and exogenous-determination beliefs.
In short, in two data sets, and in every specification and sub-sample, self- and
exogenous-determination beliefs have large and significant effects on support for
redistribution. Tax cost concerns appear not to explain this effect. A more likely
interpretation is that people are reciprocal or hold strong principles of equity. A strong
taste for equity or reciprocity is consistent with the basic concept of insuring industrious
people against bad luck, but not providing unconditional assistance to the poor if their
condition is due to idleness. Nonetheless, altruism may play a role as well, because
by lack of effort are significantly less supportive of all three types of spending. Given the “improving the
conditions” wording of the race question, the concern that spending will reduce effort and increase the
caseload seems misplaced. It seems particularly unlikely that foreign aid would cause individual transfer
disincentives. On the other hand, if people are simply unwilling to support the undeserving poor but will
support the deserving poor, then self- and exogenous-determination beliefs could have an effect on all three
of these questions.
20
N = 567.
21
N = 568
12
simply perceiving that there is more poverty increases support for it in high-income
samples.
The importance if social preferences suggests a broader search for plausible
resolutions to outstanding puzzles in the political economy of the welfare states and
taxation. For example, according to models based on the standard self-interest
assumption (Meltzer and Richard (1981), Sinn (1995)) demand for redistribution should
increase as the ratio of the median income to the mean income in the economy decreases.
However, whether or not this is the case empirically has been a matter of some debate.
Extending our understanding about what motivates support for and opposition to
redistribution may guide research on this question. The effects of inequality on demand
for redistribution differ from those of the standard model when social preferences are
important. For instance, one possibility might be that inequality increases social distance
between the rich and the poor, making the rich less sympathetic, or causing them to hold
the poor more responsible for their low incomes. Another possibility is that the effects
of income inequality on demand for redistribution are less important than other factors
such as actual and perceived racial composition of welfare recipients (Luttmer (1998)), or
design features of the redistributive programs themselves that make them more or less
likely to insure poverty caused by exogenous circumstances.
13
Table 1. Means and standard deviations from the full sample and the high
income, upwardly mobile sub-sample of the labor force.
Variable
Full Sample
n=2738
Support for redistribution (scale from 1 to 6)
U.S. is a society of haves and have-nots
Luck and effort cause poverty
Bad Luck causes poverty
Luck and effort cause wealth
Good Luck causes wealth
Plenty of Opportunity in US
10,000≤Y<15,000
15,000≤Y<20,000
20,000≤Y<30,000
30,000≤Y<50,000
50,000≤Y<75,000
75,000≤Y<100,000
100,000≤Y<150,000
150,000≤Y
Some college or more
White
Male
Age
Employed part-time
Unemployed
Married
Not a union member
Worries about bills
Most of the time
Some of the time
Almost never
-2.945
0.380a
0.122
0.405
0.099
0.336
0.828
0.037
0.059
0.148
0.299
0.225
0.103
0.059
0.034
0.617
0.798
0.509
39.229
0.176
0.020
0.554
0.878
(1.481)
(0.486)
(0.327)
(0.491)
(0.299)
(0.472)
(0.377)
(0.189)
(0.236)
(0.355)
(0.458)
(0.418)
(0.304)
(0.236)
(0.182)
(0.486)
(0.402)
(0.500)
(12.576)
(0.381)
(0.140)
(0.497)
(0.327)
0.123
0.473
0.322
(0.329)
(0.499)
(0.467)
Income≥75,000
Upwardly mobile
And
worries about bills
less often than
“All of the time”
N = 333
-3.387 (1.522)
b
0.373 (0.484)
0.117 (0.322)
0.381 (0.486)
0.093 (0.291)
0.225 (0.418)
0.916 (0.278)
0.547
0.276
0.177
0.835
0.835
0.535
41.081
0.111
0.006
0.727
0.898
(0.499)
(0.448)
(0.382)
(0.372)
(0.372)
(0.500)
(10.604)
(0.315)
(0.077)
(0.446)
(0.303)
0.048
0.417
0.535
(0.214)
(0.494)
(0.500)
n = 2760
N = 333
Why people get ahead (scale from 1 to 5)
Willingness to take risks
3.988 (0.901)
4.123 (0.784)
Inherited money
3.207 (1.279)
3.162 (1.211)
Hard work and initiative
4.661 (0.658)
4.706 (0.594)
Inborn ability and talent
3.679 (1.017)
3.667 (0.969)
Dishonesty
2.290 (1.412)
2.024c (1.243)
Good luck
3.216 (1.205)
3.180 (1.134)
Parents and family
4.449 (0.860)
4.477 (0.743)
Good looks
3.375 (1.118)
3.210 (1.011)
Connections
3.922 (1.014)
3.865 (0.923)
Race
2.825 (1.294)
2.817 (1.197)
Educ
4.586 (0.687)
4.520 (0.697)
Gender
2.861 (1.262)
2.763 (1.183)
Notes: All variables are dummy variables except “support for redistribution” and “why people get ahead.”
a
N = 2719. bN = 332. cN = 331.
14
Table 2. Ordered probit analysis of support for redistribution
Independent variable
Full Sample
Income≥75,000
N = 2738
Upwardly mobile
And
Worries about bills
less often than “All
of the time”
N = 333
Coefficients
Coefficients
(standard errors)
(standard errors)
U.S. is a society of haves and
have-nots
Luck and effort cause poverty
0.213*** (0.074)
-0.030
(0.225)
Bad luck causes poverty
0.619*** (0.047)
0.868*** (0.144)
Luck and effort cause wealth
0.341*** (0.080)
0.463*
(0.259)
Good luck causes wealth
0.444*** (0.047)
0.341** (0.152)
Plenty of Opportunity in US
-0.535*** (0.059)
-0.953*** (0.245)
10,000≤Y<15,000
-0.222
(0.156)
0.006
(0.143)
15,000≤Y<20,000
-0.146
(0.129)
20,000≤Y<30,000
-0.059
(0.125)
30,000≤Y<50,000
-0.186
(0.129)
50,000≤Y<75,000
-0.143
(0.141)
0.175
(0.168)
75,000≤Y<100,000
-0.135
(0.151)
0.057
(0.182)
100,000≤Y<150,000
-0.338** (0.167)
-0.230*** (0.047)
-0.360* (0.163)
150,000≤Y
-0.240***
(0.053)
-0.354* (0.168)
Some college or more
-0.304***
(0.045)
-0.231** (0.125)
White
0.010
(0.009)
-0.004
(0.038)
Male
0.000
(0.000)
0.000
(0.000)
Age
2
0.088
(0.059)
0.082
(0.202)
Age
-0.021
(0.151)
-0.178
(0.900)
Employed part-time
-0.083* (0.045)
-0.161
(0.148)
Unemployed
-0.219*** (0.064)
-0.002
(0.201)
Married
Not a union member
-0.048
(0.093)
Worries about bills
-0.138* (0.080)
-0.044
(0.283)
Most of the time
-0.251*** (0.085)
-0.104
(0.282)
Some of the time
Almost never
Pseudo R2
.092
.110
Income≥75,000
Upwardly mobile
And
worries about bills
less often than “All
of the time”
N = 332
Coefficients
(standard errors)
0.436*** (0.133)
0.014
0.830***
0.422
0.264*
-0.904***
(0.226)
(0.146)
(0.260)
(0.155)
(0.246)
0.228
0.084
(0.169)
(0.183)
-0.374**
-0.367**
-0.188
-0.003
0.000
0.128
-0.300
-0.079
-0.011
(0.164)
(0.168)
(0.126)
(0.038)
(0.000)
(0.203)
(0.890)
(0.150)
(0.202)
-0.023
-0.018
(0.284)
(0.126)
.120
Likelihood Ratio Test that all
self-interest coefficients = 0
Chi2(26) = 221.94
Pr > chi2 = 0.000
Chi2(19) = 25.05
Pr > chi2 = 0.159
Chi2(19) = 24.32
Pr > chi2 = 0.184
Likelihood Ratio Test that all
self- and exogenousdetermination beliefs
coefficients = 0
Chi2(5) = 505.97
Pr > chi2 = 0.000
Chi2(5) = 79.07
Pr > chi2 = 0.000
Chi2(5) = 65.34
Pr > chi2 = 0.000
Notes: Equations also include seven occupation dummies, and a dummy for household size greater than 4.
Omitted category for the “Worries about bills” is “All of the time.”
***Significant at the 1 percent level. **Significant at the 5 percent level. * Significant at the 10 percent level.
15
Table 3. Selected Marginal Effects of Regressors in Table 2, Column 1.
P[support=1] P[support=2] P[support=3] P[support=4] P[support=5]
Lack of effort causes
0.054
0.146
0.223
0.258
0.219
poverty
Bad luck causes
0.013
0.059
0.136
0.233
0.306
poverty
Change
-0.041
-0.087
-0.087
-0.025
0.087
Strong effort causes
0.045
0.131
0.212
0.26
0.236
wealth
Good luck causes
0.016
0.068
0.148
0.242
0.299
wealth
Change
-0.029
-0.063
-0.064
-0.018
0.063
Not much
0.010
0.05
0.121
0.222
0.313
opportunity in U.S.
Plenty of
0.037
0.116
0.2
0.26
0.252
opportunity in U.S.
Change
0.027
0.066
0.079
0.038
-0.061
Non-white
0.019
0.076
0.159
0.248
0.292
White
0.033
0.109
0.194
0.26
0.259
Change
0.014
0.033
0.035
0.012
-0.033
Female
0.021
0.081
0.164
0.25
0.288
Male
0.042
0.125
0.208
0.261
0.242
Change
0.021
0.044
0.044
0.011
-0.046
No college
0.022
0.083
0.166
0.251
0.286
Some college or
0.036
0.116
0.199
0.26
0.252
more
Change
0.014
0.033
0.033
0.009
-0.034
Income<10,000
0.023
0.085
0.169
0.252
0.284
Income≥150,000
0.048
0.136
0.216
0.26
0.23
Change
0.025
0.051
0.047
0.008
-0.054
Marginal effects are in bold type in the rows labeled “Change.” The marginal effects shown are the
differences between two types of respondent (e.g., those who reported that bad luck causes poverty
relative to those who reported that lack of effort causes poverty) in the probabilities of having a
particular value of the dependent variable support for redistribution.
16
P[support=6]
0.099
0.252
0.153
0.116
0.227
0.111
0.285
0.135
-0.150
0.206
0.149
-0.057
0.196
0.123
-0.073
0.192
0.136
-0.056
0.187
0.110
-0.077
Table 4. Ordered probit model testing the interaction effects
between race and gender and beliefs that race and gender are
important to getting ahead in life.
Dependent variable: support for redistribution
Independent variable
Why people get ahead (Scale from 1 to 5):
Willingness to take risks
Inherited money
Inborn ability and talent
Dishonesty
Parents and family
Good looks
Connections
Educ
Hard work and initiative
Luck
Race
Race*respondent is white
Gender
Gender*respondent is male
Respondent is white
Respondent is male
-0.091***
0.057***
-0.002
0.043***
0.064***
-0.047**
0.064***
0.060*
-0.157***
0.040**
0.043
0.078**
0.061**
-0.007
-0.438***
-0.309***
Pseudo R2
N = 2760
(0.023)
(0.018)
(0.021)
(0.015)
(0.025)
(0.020)
(0.023)
(0.032)
(0.034)
(0.019)
(0.032)
(0.036)
(0.025)
(0.032)
(0.122)
(0.103)
.063
Notes: This specification also includes all of the objective and subjective socio-economic proxies included in
the specifications of Table 2.
***
Significant at the 1% level. **Significant at the 5% level. * Significant at the 10% level.
17
Figure 1.
Estimated Probabilities for Four Categories of People
Full Sample (N = 2738)
0.6
Estimated Probabilitiy
0.5
0.4
Believes in Luck
Believes in Effort
0.3
Low Socio-Econ Status
High Socio-Econ Status
0.2
0.1
0
1
2
3
4
Support for Redistribution
18
5
6
I am especially indebted to Samuel Bowles for his guidance in this research, and am
grateful to two anonymous referees for comments and George Loewenstein for helpful
discussion. Financial support was provided by the MacArthur Foundation Network on
Inequality, Washington University, and the MacArthur Foundation Preferences Network.
References
Andreoni, J., 1989. Giving with Impure Altruism: Applications to Charity and Ricardian
Equivalence. Journal of Political Economy 97(6), 1447-1458.
Barr, N., 1992. Economic Theory and the Welfare State: A Survey and Interpretation.
Journal of Economic Literature. 30 (June), 741-803.
Blount, S., 1995. When Social Outcomes Aren’t Fair: The Effect of Causal Attributions
on Preferences. Organizational Behavior and Human Decision Processes 63 (2), 131144.
Bowles, S., Gintis, H., 2000. Reciprocity, Self-Interest and the Welfare State. The
Nordic Journal of Political Economy. 26 (Jan.).
Deutsch, M., 1985. Distributive Justice. Yale University Press, New Haven.
Gallup Organization, 1998. Haves and Have-Nots: Perceptions of Fairness and
Opportunity. Gallup Poll Social Audit Survey conducted by telephone with a
randomly selected national sample of 5001 adults conducted between April 23 and
May 31, 1998.
Gilens, M. 1999. Why Americans Hate Welfare: Race, Media, and the Politics of AntiPoverty Policy. Chicago University Press.
Greene, W.H., 1997. Econometric Analysis. MacMillan, New York.
Kluegel, J., Smith, E., 1986. Beliefs About Inequality: Americans Views of What is and
What Ought to Be. Aldine de Gruyter, New York.
Luttmer, E., 1998. Group Loyalty and the Taste for Redistribution. Mimeographed. The
Irving B. Harris School of Public Policy Studies, The University of Chicago.
Meltzer, A., Richard, S, 1981. A Rational Theory of the Size of Government. Journal of
Political Economy 89 (5), 814-927.
Miller, D., 1992. Distributive Justice: What the People Think. Ethics. 102 (April), 555593.
Moffitt, R., 1998. The decline of Welfare Benefits in the U.S.: The Role of Wage
Inequality. Journal of Public Economics. 68 (3), 421-452.
Moffitt, R., 1999. Explaining Welfare Reform: Public Choice and the Labor Market.
Mimeographed. Johns Hopkins University.
Piketty, T., 1995. Social Mobility and Redistributive Politics. The Quarterly Journal of
Economics CX (3), 551-584.
Piketty, T., 1999. Attitudes Toward Income Inequality in France: Do People Really
Disagree? Mimeographed, Cepremap, Paris.
Rabin, M., 1993. Incorporating Fairness into Game Theory. American Economic Review
83 (5), 1281-1302.
Romer, Paul 1994. Preferences, Promises, and the Politics of Entitlement. In:
19
Fuchs, V.R., 1994. Individual and Social Responsibility: Child Care, Education, Medical
Care, and Long-Term Care in America. The University of Chicago Press, Chicago and
New York, pp. 195-228.
Sinn, H.W., 1995. A Theory of the Welfare State. Scandinavian Journal of Economics, 97:
495-526.
Sugden, R., 1982. On the Economics of Philanthropy. Economic Journal, 92 (June): 341350.
Walster, E.G., Walster, W., Berscheid, E. 1978. Equity: Theory and Research. Allyn and
Bacon, Boston.
20
Appendix
Dependent Variables in Gallup Data
The measure of redistributive preferences is a summative scale of the questions
below. The measure is increasing in support for redistribution, so the signs of questions
1, 2, 3, and 5 are reversed.
Table A1.
Questions used in the construction of the dependent variable measuring support for redistribution.
Question
Aggregate Response (N=2738)*
1) People feel differently about how far a
Should: 46.1%
government should go. Here is a phrase which
Should Not: 53.9%
some people believe in and some don’t. Do you
think our government should or should not
redistribute wealth by heavy taxes on the rich?
2) Some people feel that the government in
Government should help the poor: 69.7%
Washington, DC should make every possible effort The poor should help themselves: 30.3%
to improve the social and economic position of the
poor. Others feel that the government should not
make any special effort to help the poor, because
they should help themselves. How do you feel
about this?
3) Which one of the following groups do you think Groups other than the poor: 69.8%
has the greatest responsibility for helping the poor:
The poor themselves: 30.2%
churches, private charities, the government, the
families and relatives of poor people, the poor
themselves, or someone else?
4) Do you feel that the distribution of money and
Distribution is fair: 33.4%
wealth in this country today is fair, or do you feel
Should be more evenly distributed: 66.6%
that the money and wealth in this country should be
more evenly distributed among a larger percentage
of the people?
5) Do you think that the fact that some people in
Problem: 53.3%
the United States are rich and others are poor 1)
Acceptable: 46.8%
represents a problem that needs to be fixed or 2) Is
an acceptable part of our economic system?
Note: Of respondents with incomes of at least $150,000 and who also expected upward mobility in the
next five years, the percentage responding in favor of redistribution is 24%, 67%, 63%, 48%, and 45% for
questions 1 through 5, respectively. In this sub-sample, there are 84 observations for all but the second
question, which has 81 observations.
21
Independent Variables in Gallup Data
Plenty of Opportunity in the US: Some people say that there’s not much opportunity in
America today – that the average person doesn’t have much chance to really get ahead.
Others say there’s plenty of opportunity and anyone who works hard can go as far as
they want. Which one comes closer to the way you feel about this? 1) Not much
opportunity 2) Plenty of opportunity
Causes of poverty: Just in your opinion, which is more often to blame if a person is poor
– lack of effort on his or her part, or circumstances beyond his or her control? 1) Lack of
effort 2) Both 3) Luck or circumstances beyond his/her control.
Causes of wealth: Just in your opinion, which is more often to blame if a person is rich –
strong effort on his or her part, or circumstances beyond his or her control? 1) Strong
effort 2) Both 3) Luck or circumstances beyond his/her control.
Determinants of Success: I am going to read several reasons why some people get ahead
and succeed in life and others do not. Using a one-to-five scale, where “1” means not at
all important and “5” means extremely important, please tell me how important it is as a
reason for a person’s success. You can choose any number from one to five. First,
A: How important is willingness to take risks
B: How important is money inherited from families
C: How important is hard work and initiative
D: How important is ability or talent that a person is born with
E: How important is dishonesty and willingness to take what they can get
F: How important is good luck, being in the right place at the right time
G: How important is parents and the family environment they grow up in
H: How important is physical appearance and good looks
I: How important is [sic] connections and knowing the right people
J: How important is being a member of a particular race or ethnic group
K: How important is getting the right education or training
L: How important is a person’s gender, that is whether they are male or female
America is society of haves and have-nots: Some people think of American society as
divided into two groups – the “haves” and “have-nots”, while others think it’s incorrect
to think of America that way. Do you, yourself, think of America as divided into haves
and have-nots, or don’t you think of America that way? 1) Yes, think that way 2) No,
don’t think that way.
Mobility: Mobility measures are constructed from the following three questions:
1. “Think of a picture of a ladder. Suppose we say that the top of the ladder
represents the best possible life for you, and the bottom represents the worst
22
possible life for you. If the top step is “10” and the bottom step is “0”, on
which step of the ladder do you feel you personally stand at the present time?
2. “On which step would you say you stood five years ago?”
3. “Just your best guess, on which step do you think you will stand in the
future, say about five years from now?”
Past mobility is the current ladder position minus the past ladder position. Future
mobility is the future ladder position minus the current ladder position.
Variables in General Social Survey Data
“We are faced with many problems in our country, none of which can be solved easily of
inexpensively. I’m going to name some of these problems, and for each one I’d like you
to tell me whether you think we’re spending too much money on it, too little money, or
about the right amount. First (READ ITEM A) … are we spending too much, too little,
or about the right amount on (ITEM)”
Items Used:
Halting the rising crime rate: 1) too little 2) about right 3) too much
Dealing with drug addiction: 1) too little 2) about right 3) too much
Improving the conditions of blacks: 1) too little 2) about right 3) too much
The military, armaments and defense: 1) too little 2) about right 3) too much
Foreign aid: 1) too little 2) about right 3) too much
Welfare: 1) too little 2) about right 3) too much
Note: “Don’t know” is coded as missing. I reverse the order of the responses so
that they increase in support for spending.
“Now I will [read] a list of reasons some people give to explain why there are poor
people in this country. Please tell me whether you feel each of these is very important,
somewhat important, or not important in explaining why there are poor people in this
country.”
Lack of effort by the poor themselves: 1) very important 2) somewhat important
3) not important
Note: “Don’t know” is coded as missing. I reverse the order of the responses so
that the measure increases in beliefs in self-determination.
23